Cognition does not affect perception: Evaluating the evidence for … · 2017. 5. 15. · Cognition does not affect perception: Evaluating the evidence for “top-down” effects

Cognition does not affect perception:Evaluating the evidence for “top-down”effects

Chaz FirestoneDepartment of Psychology, Yale University, New Haven, CT 06520-8205

[email protected]

Brian J. SchollDepartment of Psychology, Yale University, New Haven, CT [email protected]

Abstract:What determines what we see? In contrast to the traditional “modular” understanding of perception, according to which visualprocessing is encapsulated from higher-level cognition, a tidal wave of recent research alleges that states such as beliefs, desires, emotions,motivations, intentions, and linguistic representations exert direct, top-down influences on what we see. There is a growing consensus thatsuch effects are ubiquitous, and that the distinction between perception and cognition may itself be unsustainable. We argue otherwise:None of these hundreds of studies – either individually or collectively – provides compelling evidence for true top-down effects onperception, or “cognitive penetrability.” In particular, and despite their variety, we suggest that these studies all fall prey to only ahandful of pitfalls. And whereas abstract theoretical challenges have failed to resolve this debate in the past, our presentation of thesepitfalls is empirically anchored: In each case, we show not only how certain studies could be susceptible to the pitfall (in principle),but also how several alleged top-down effects actually are explained by the pitfall (in practice). Moreover, these pitfalls are perfectlygeneral, with each applying to dozens of other top-down effects. We conclude by extracting the lessons provided by these pitfalls intoa checklist that future work could use to convincingly demonstrate top-down effects on visual perception. The discovery ofsubstantive top-down effects of cognition on perception would revolutionize our understanding of how the mind is organized; butwithout addressing these pitfalls, no such empirical report will license such exciting conclusions.

1. Introduction

How does the mind work? Though this is, of course, thecentral question posed by cognitive science, one of thedeepest insights of the last half-century is that the questiondoes not have a single answer: There is no one way themind works, because the mind is not one thing. Instead,the mind has parts, and the different parts of the mindoperate in different ways: Seeing a color works differentlythan planning a vacation, which works differently thanunderstanding a sentence, moving a limb, remembering afact, or feeling an emotion.

The challenge of understanding the natural world is tocapture generalizations – to “carve nature at its joints.”Where are the joints of the mind? Easily, the mostnatural and robust distinction between types of mentalprocesses is that between perception and cognition. Thisdistinction is woven so deeply into cognitive science asto structure introductory courses and textbooks, differen-tiate scholarly journals, and organize academic depart-ments. It is also a distinction respected by commonsense: Anyone can appreciate the difference between,on the one hand, seeing a red apple and, on the otherhand, thinking about, remembering, or desiring a redapple. This difference is especially clear when perception

and cognition deliver conflicting evidence about theworld – as in most visual illusions. Indeed, there may beno better way to truly feel the distinction between percep-tion and cognition for yourself than to visually experiencethe world in a way you know it not to be.There is a deep sense in which we all know what per-

ception is because of our direct phenomenologicalacquaintance with percepts – the colors, shapes, andsizes (etc.) of the objects and surfaces that populateour visual experiences. Just imagine looking at anapple in a supermarket and appreciating its redness (asopposed, say, to its price). That is perception. Or lookat Figure 1A and notice the difference in lightnessbetween the two gray rectangles. That is perception.Throughout this paper, we refer to visual processingsimply as the mental activity that creates such sensations;we refer to percepts as the experiences themselves, andwe use perception (and, less formally, seeing) to encom-pass both (typically unconscious) visual processing andthe (conscious) percepts that result.

1.1. The new top-down challenge

Despite the explanatorily powerful and deeply intuitivenature of the distinction between seeing and thinking, a

BEHAVIORAL AND BRAIN SCIENCES (2016), Page 1 of 77doi:10.1017/S0140525X15000965, e229

© Cambridge University Press 2016 0140-525X/16 1https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0140525X15002800Downloaded from https:/www.cambridge.org/core. NYU Medical Center: Ehrman Medical Library, on 15 May 2017 at 21:10:16, subject to the Cambridge Core terms of use, available at

mailto:[email protected]:[email protected]:/www.cambridge.org/core/termshttps://doi.org/10.1017/S0140525X15002800https:/www.cambridge.org/core

vocal chorus has recently and vigorously challenged theextent of this division, calling for a generous blurring ofthe lines between visual perception and cognition (forrecent reviews, see Balcetis 2016; Collins & Olson 2014;Dunning & Balcetis 2013; Goldstone et al. 2015; Lupyan2012; Proffitt & Linkenauger 2013; Riccio et al. 2013; Ste-fanucci et al. 2011; Vetter & Newen 2014; Zadra & Clore2011). On this increasingly popular view, higher-level cog-

nitive states routinely “penetrate” perception, such thatwhat we see is an alloy both of bottom-up factors and ofbeliefs, desires, motivations, linguistic representations,and other such states. In other words, these views holdthat the mental processes responsible for building perceptscan and do access radically more information elsewhere inthe mind than has traditionally been imagined.At the center of this dispute over the nature of visual per-

ception and its relation to other processes in the mind hasbeen the recent and vigorous proliferation of so-called top-down effects on perception. In such cases, some extraper-ceptual state is said to literally and directly alter what wesee. (As of this writing, we count more than 175 paperspublished since 1995 reporting such effects; for a list,see http://perception.yale.edu/TopDownPapers.) For example,it has been reported that desiring an object makes it lookcloser (Balcetis & Dunning 2010), that reflecting on uneth-ical actions makes the world look darker (Banerjee et al.2012), that wearing a heavy backpack makes hills looksteeper (Bhalla & Proffitt 1999), that words having to dowith morality are easier to see (Gantman & Van Bavel2014), and that racial categorization alters the perceivedlightness of faces (Levin & Banaji 2006).If what we think, desire, or intend (etc.) can affect

what we see in these ways, then a genuine revolution inour understanding of perception is in order. Notice, forexample, that the vast majority of models in visionscience do not consider such factors; yet, apparently,such models have been successful! For example, today’svision science has essentially worked out how low-levelcomplex motion is perceived and processed by thebrain, with elegant models of such processes accountingfor extraordinary proportions of variance in motion pro-cessing (e.g., Rust et al. 2006) – and this success hascome without factoring in morality, hunger, or language(etc.). Similarly, such factors are entirely missing fromcontemporary vision science textbooks (e.g., Blake &Sekuler 2005; Howard & Rogers 2002; Yantis 2013). If

CHAZ FIRESTONE is a graduate student in the Depart-ment of Psychology at Yale University. As of July2017, he will be an Assistant Professor in the Depart-ment of Psychological and Brain Sciences at JohnsHopkins University. He holds an Sc.B. in cognitive neu-roscience and an A.M. in philosophy, both from BrownUniversity. His research explores the border betweenperception and cognition, and he was recognized forthis work with the 2013 William James Prize from theSociety for Philosophy and Psychology. He hasn’t pub-lished very many papers, but this one is his favorite.

BRIAN SCHOLL is Professor of Psychology and Chair ofthe Cognitive Science program at Yale University,where he also directs the Perception & CognitionLaboratory. He and his research group have publishedmore than 100 papers on various topics in cognitivescience, with a special focus on how visual perceptioninteracts with the rest of the mind. He is a recipientof the Distinguished Scientific Award for Early CareerContribution to Psychology and the Robert L. FantzMemorial Award, both from the American Psychologi-cal Association, and is a past President of the Societyfor Philosophy and Psychology. At Yale he is a recipientof both the Graduate Mentor Award and the Lex HixonPrize for Teaching Excellence in the Social Sciences,and he has great fun teaching the Introduction toCognitive Science course.

A CB

D E

Figure 1. Examples of lightness illusions can be subjectively appreciated as “demonstrations” (for references and explanations, seeAdelson 2000). (A) The two columns of gray rectangles have the same luminance, but the left one looks lighter. (B) The rectanglesare uniformly gray, but they appear to lighten and darken along their edges. (C) Uniformly colored squares of increasing luminanceproduce an illusory light “X” shape at their corners. (D) The two central squares have the same objective luminance, but the left onelooks lighter. (E) The two rectangles are identical segments of the same gradient, but the right one looks lighter. Similardemonstrations abound, for nearly every visual feature.

Firestone and Scholl: Cognition does not affect perception

2 BEHAVIORAL AND BRAIN SCIENCES, 39 (2016)https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0140525X15002800Downloaded from https:/www.cambridge.org/core. NYU Medical Center: Ehrman Medical Library, on 15 May 2017 at 21:10:16, subject to the Cambridge Core terms of use, available at

http://perception.yale.edu/TopDownPapershttps:/www.cambridge.org/core/termshttps://doi.org/10.1017/S0140525X15002800https:/www.cambridge.org/core

such factors do influence how we see, then such modelsand textbooks are scandalously incomplete.

Although such factors as morality, hunger, and languageare largely absent from contemporary vision science inpractice, the emergence of so many empirical papersreporting top-down effects of cognition on perception hasshifted the broader consensus in cognitive science.Indeed, such alleged top-down effects have led severalauthors to declare that the revolution in our understandingof perception has already occurred, proclaiming as dead notonly a “modular” perspective on vision, but often also thevery distinction between perception and cognition itself.For example, it has been asserted that it is a “generallyaccepted concept that people tend to see what they wantto see” (Radel & Clément-Guillotin 2012, p. 233), andthat “the postulation of the existence of visual processesbeing functionally encapsulated…cannot be justifiedanymore” (Vetter & Newen 2014, p. 73). This sort of evi-dence led one philosopher to declare, in an especiallysweeping claim, that “[a]ll this makes the lines betweenperception and cognition fuzzy, perhaps even vanishing”and to deny that there is “any real distinction between per-ception and belief” (Clark 2013, p. 190).

1.2. Our thesis and approach

Against this wealth of evidence and its associated consen-sus, the thesis of this paper is that there is in fact no evi-dence for such top-down effects of cognition on visualperception, in every sense these claims intend. With hun-dreds of reported top-down effects, this is, admittedly, anambitious claim. Our aim in this discussion is thus to explic-itly identify the (surprisingly few, and theoretically interest-ing) “pitfalls” that account for reports of top-downpenetration of visual perception without licensing suchconclusions.

Our project differs from previous theoretical challenges(e.g., Fodor 1984; Pylyshyn 1999; Raftopoulos 2001a) inseveral ways. First, whereas many previous discussionsdefended the modular nature of only a circumscribed(and possibly unconscious) portion of visual processing(e.g., “early vision”; Pylyshyn 1999), we have the broaderaim of evaluating the evidence for top-down effects onwhat we see as a whole – including visual processing andthe conscious percepts it produces. Second, several pitfallswe present are novel contributions to this debate. Third,and most important, whereas past abstract discussionshave failed to resolve this debate, our presentation ofthese pitfalls is empirically anchored: In each case, weshow not only how certain studies could be susceptible tothe pitfall (in principle), but also how several alleged top-down effects actually are explained by the pitfall (in prac-tice, drawing on recent and decisive empirical studies).Moreover, each pitfall we present is perfectly general,applying to dozens more reported top-down effects.Research on top-down effects on visual perception musttherefore take the pitfalls seriously before claims of suchphenomena can be compelling.

The question of whether there are top-down effects ofcognition on visual perception is one of the most founda-tional questions that can be asked about what perceptionis and how it works, and it is therefore no surprise thatthe issue has been of tremendous interest (especiallyrecently) – not only in all corners of psychology, but also

in neighboring disciplines such as philosophy of mind(e.g., Macpherson 2012; Siegel 2012), neuroscience (e.g.,Bannert & Bartels 2013; Landau et al. 2010), psychiatry(e.g., Bubl et al. 2010), and even aesthetics (e.g., Nanay2014; Stokes 2014). It would be enormously exciting to dis-cover that perception changes the way it operates in directresponse to goings-on elsewhere in the mind. Our hope isthus to help advance future work on this foundational ques-tion, by identifying and highlighting the key empiricalchallenges.

2. A recipe for revolution

The term top-down is used in a spectacular variety of waysacross many literatures. What do we mean when we say thatcognition does not affect perception, such that there are notop-down effects on what we see? The primary reasonthese issues have received so much historical and contem-porary attention is that a proper understanding of mentalorganization depends on whether there is a salient “joint”between perception and cognition. Accordingly, we focuson the sense of top-down that directly addresses thisaspect of how the mind is organized. This sense of theterm is, for us, related to traditional questions of whethervisual perception is modular, encapsulated from the restof cognition, and “cognitively (im)penetrable.”1 At issue isthe extent to which what and how we see is functionallyindependent from what and how we think, know, desire,act, and so forth. We single out this meaning of top-downnot only because it may be the most prominent usage ofthe term, but also because the questions it raises are espe-cially foundational for our understanding of the organiza-tion of the mind.Nevertheless, there are several independent uses of top-

down that are less revolutionary and do not directly interactwith these questions.

2.1. Changing the processing versus (merely) changingthe input

On an especially permissive reading of “top-down,” top-down effects are all around us, and it would be absurd todeny cognitive effects on what we see. For example,there is a trivial sense in which we all can willfully controlwhat we visually experience, by (say) choosing to closeour eyes (or turn off the lights) if we wish to experiencedarkness. Though this is certainly a case of cognition (spe-cifically, of desire and intention) changing perception, thisfamiliar “top-down” effect clearly isn’t revolutionary,insofar as it has no implications for how the mind is orga-nized – and for an obvious reason: Closing your eyes (orturning off the lights) changes only the input to perception;it does not change perceptual processing itself.Despite the triviality of this example, the distinction is

worth keeping in mind, because it is not always obviouswhen an effect operates by changing the input. To takeone fascinating example, facial expressions associated withfear (e.g., widened eyes) and disgust (e.g., narrowed eyes)have recently been shown to reliably vary the eye-aperturediameter, directly influencing acuity and sensitivity by alter-ing the actual optical information reaching perceptual pro-cessing (Lee et al. 2014). (As we will see later, the


BEHAVIORAL AND BRAIN SCIENCES, 39 (2016) 3https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0140525X15002800Downloaded from https:/www.cambridge.org/core. NYU Medical Center: Ehrman Medical Library, on 15 May 2017 at 21:10:16, subject to the Cambridge Core terms of use, available at

https:/www.cambridge.org/core/termshttps://doi.org/10.1017/S0140525X15002800https:/www.cambridge.org/core

distinction between input and processing also arises withregard to perceptual vs. attentional effects.)

2.2. Descending neural pathways

In systems neuroscience, some early models of brain func-tion were largely feedforward, with various brain regionsfeeding information to each other in a unidirectionalsequence. In contrast, there is now considerable evidencethat brain regions that were initially considered “higherup” in a processing hierarchy can modulate “lower”regions, through so-called re-entrant processing fromdescending neural pathways – and these sorts of modula-tion are often also commonly called top-down effects(e.g., Gilbert & Li 2013; Rolls 2008; Zhang et al. 2014).Though extremely interesting for certain questions aboutfunctional neuroanatomy, this type of “top-down” influencehas no necessary implications for cognitive penetrability.One reason is that nearly all brain regions subserve multiplefunctions. Even parts of visual cortex, for example, areinvolved in imagery (e.g., Kosslyn 2005), recall (e.g., LeBihan et al. 1993), and reward processing (Vickery et al.2011) – so that it is almost never clear which mentalprocess a descending pathway is descending to (or if thatdescending pathway is influencing the input or the process-ing of whatever it descends to, per sect. 2.1).At any rate, we do not discuss descending pathways in

the brain in this target article, for two reasons. First, theimplications of this body of work for issues of modularityand cognitive penetrability have been addressed and cri-tiqued extensively elsewhere (e.g., Raftopoulos 2001b).Second, our aim here is to focus on that recent wave ofwork that promises a revolution in how we think aboutthe organization of the mind. And whatever one thinks ofthe relevance of descending neural pathways to issues ofwhether cognition affects perception, they certainlycannot be revolutionary today: The existence of descendingneural pathways has been conclusively established manytimes over, and they are now firmly part of the orthodoxyin our understanding of neural systems.

2.3. Top-down effects versus context effects and“unconscious inferences” in vision

Visual processing is often said to involve “problem solving”(Rock 1983) or “unconscious inference” (Gregory 1980;Helmholtz 1866/1925). Sometimes these labels areapplied to seemingly sophisticated processing, as inresearch on the perception of causality (e.g., Rolfs et al.2013; Scholl & Tremoulet 2000) or animacy (e.g., Gaoet al. 2010; Scholl & Gao 2013). But more often, thelabels are applied to relatively early and low-level visualprocessing, as in the perception of lightness (e.g.,Adelson 2000) or depth (e.g., Ramachandran 1988). Inthose cases, such terminology (which may otherwiseevoke notions of cognitive penetrability) refers to aspectsof processing that are wired into the visual module itself(so-called “natural constraints”) – and so do not at allimply effects of cognition on perception, or “top-down”effects. This is true even when such processing involvescontext effects, wherein perception of an object may beinfluenced by properties of other objects nearby (e.g., asin several of the lightness illusions in Fig. 1). In suchcases, the underlying processes continue to operate

reflexively (based solely on their visual input) regardlessof your cognitive inferences or problem-solving strategies(for discussion, see Scholl & Gao 2013) – as when lightnessillusions based on “unconscious inferences” persist in theface of countervailing knowledge (Fig. 1). (For further dis-cussion of why vision being “smart” in such ways does notimply cognitive penetrability, see Kanizsa 1985; Pylyshyn1999.)

2.4. Cross-modal effects

What we see is sometimes affected by other sense modali-ties. For example, a single flash of light can appear to flickerwhen accompanied by multiple auditory beeps (Shamset al. 2000), and two moving discs that momentarilyoverlap are seen to bounce off each other (rather thanstream past each other) if a beep is heard at the momentof overlap (Sekuler et al. 1997). However, these cases –though interesting for many other reasons – do not demon-strate cognitive penetrability, for much the same reasonthat unconscious inferences in vision fail to do so. Forexample, such crossmodal integration is itself a reflexive,apparently impenetrable process: The sounds’ effectsoccur “whether you like it or not,” and they occur extremelyquickly (e.g., in less than 100 ms; Shams et al. 2002). Col-lectively, such results are consistent with the entireprocess being contained within perception itself, ratherthan being an effect of more central cognitive processeson perception.At any rate, we do not discuss crossmodal effects here. As

with descending neural pathways, whatever one thinks ofthe relevance of this work to the issues we discuss, they cer-tainly cannot be revolutionary today in the way promised bythe work we review in section 3 – if only because the exis-tence of crossmodal effects has been conclusively estab-lished and is common ground for all parties in thisdiscussion.

2.5. Input-driven changes in sensitivity over time

Despite encapsulation, input may sometimes change visualprocessing by increasing sensitivity over time to certainvisual features. For example, figure–ground assignmentfor ambiguous stimuli is sometimes biased by experience:The visual system will more likely assign figure to familiarshapes, such as the profile of a woman with a skirt (Peterson& Gibson 1993; Fig. 2A). However, such changes don’tinvolve any penetration because they don’t involve effectsof knowledge per se. For example, inversion eliminatesthis effect even when subjects know the inverted shape’sidentity (Peterson & Gibson 1994). Therefore, what maysuperficially appear to be an influence of knowledge on per-ception is simply increased sensitivity to certain contours.Indeed, Peterson and Gibson (1994) volunteer that theirphenomena don’t reflect top-down effects, and in particu-lar that “the orientation dependence of our results demon-strates that our phenomena are not dependent on semanticknowledge” (p. 561). Thus, such effects aren’t “top-down”in any sense that implies cognitive penetrability, becausethe would-be penetrator is just the low-level visual inputitself. (Put more generally, the thesis of cognitive impene-trability constrains the information modules can access, butit does not constrain what modules can do with the inputthey do receive; e.g., Scholl & Leslie 1999.)




3. Contemporary top-down effects

What remains after setting aside alternative meanings of“top-down effects” is the provocative claim that ourbeliefs, desires, emotions, actions, and even the languageswe speak can directly influence what we see. Much inkhas been spilled arguing whether this should or shouldn’tbe true, based primarily on various theoretical consider-ations (e.g., Churchland 1988; Churchland et al. 1994;Firestone 2013a; Fodor 1983; 1984; 1988; Goldstone &Barsalou 1998; Lupyan 2012; Machery 2015; Proffitt &Linkenauger 2013; Pylyshyn 1999; Raftopoulos 2001b;Vetter & Newen 2014; Zeimbekis & Raftopoulos 2015).We will not engage those arguments directly – largely, weadmit, out of pessimism that such arguments can be (orhave been) decisive. Instead, our focus will be on the

nature and strength of the empirical evidence for cognitivepenetrability in practice.Though recent years have seen an unparalleled prolifera-

tion of alleged top-down effects, such demonstrations have along and storied history. One especially visible landmark inthis respect was the publication in 1947 of Bruner andGoodman’s “Value and need as organizing factors in percep-tion.” Bruner and Goodman’s pioneering study reportedthat children perceived coins as larger than they perceivedworthless cardboard discs of the same physical size, andalso that children from poor families perceived the coinsas larger than did wealthy children. These early resultsignited the New Look movement in perceptual psychology,triggering countless studies purporting to show all manner oftop-down influences on perception (for a review, see Bruner1957). It was claimed, for example, that hunger biased the

E

B 2 1 34

5

6 ?

NH

QK

blame

moralherooo

aaeeeeeame

oralero

C

TTLE6 8 9

J45%

D

L

A

RF

cleanclean

candycandylovelove

G

spiderspider

soursourcancercancer

I

M

O

P $25 $0

Figure 2. Diagrams or depictions of various possible top-down effects on perception: (A) Figure–ground assignment is biased towardfamiliar shapes, such as the profile of a woman (Peterson & Gibson 1993). (B) Being thirsty (as a result of eating salty pretzels) makesambiguous surfaces look more transparent (Changizi & Hall 2001). (C) Morally relevant words are easier to see than morally irrelevantwords (Gantman & Van Bavel 2014). (D) Wearing a heavy backpack makes hills look steeper (Bhalla & Proffitt 1999). (E) Holding a widepole makes apertures look narrower (Stefanucci & Geuss 2009). (F) Accuracy in dart throwing biases subsequent estimates of the target’ssize (Cañal-Bruland et al. 2010; Wesp & Gasper 2004). (G) Positive words are seen as lighter than negative words (Meier et al. 2007). (H)Scary music makes ambiguous images take on their scarier interpretation (Prinz & Seidel 2012). (I) Smiling faces appear brighter (Songet al. 2012). (J) Learning color–letter associations makes identically hued numbers and letters appear to have their categories’ hues (e.g.,the E will look red and the 6 will look blue, even though they are equally violet; Goldstone 1995). (K) A grayscale banana appears yellow(Hansen et al. 2006). (L) Conceptual similarity enhances size-contrast illusions (Coren & Enns 1993). (M) Labeling certain blocky figuresas “2” and “5”makes them easier to find in a visual search array (Lupyan & Spivey 2008). (N) Calligraphic knowledge (e.g., of the directionof the sixth stroke of a Chinese character) affects the direction of apparent motion when that stroke is flashed (Tse & Cavanagh 2000). (O)Reflecting on unethical actions makes the world look darker (Banerjee et al. 2012). (P) Desired objects are seen as closer, as measured bybeanbag throws (Balcetis & Dunning 2010). (Q) The middle traffic light is called gelb (yellow) in German and oranje (orange) in Dutch,which influences its perceived color (Mitterer et al. 2009). (R) You may be able to intentionally decide which interpretation of a Neckercube to see (cf. Long & Toppino 2004).




visual interpretation of ambiguous images (Lazarus et al.1953), that knowledge of objects’ typical colors influencedonline color perception (Bruner et al. 1951), and that mean-ingful religious iconography dominated other symbols inbinocular rivalry (Lo Sciuto & Hartley 1963).However, the New Look movement’s momentum even-

tually stalled as its findings buckled under methodologicaland theoretical scrutiny. For example, follow-up studieson the value-based size-distortion effects could replicatethem only when subjects made judgments from memoryrather than during online viewing (Carter & Schooler1949; see also Landis et al. 1966), and other critiques iden-tified theoretically puzzling moderating variables orreported that many other valuable objects and symbolsfailed to produce similar results (e.g., Klein et al. 1951;McCurdy 1956). Other confounding variables were eventu-ally implicated in the original effects, leading severalresearchers to conclude that “[o]nly when better experi-ments have been carried out will we be able to determinewhat portion of the effect is due to nonperceptualfactors” (Landis et al. 1966, p. 729). By the next decade,the excitement surrounding such ideas had fizzled, and“the word ‘artifact’ became the descriptive term par excel-lence associated with the New Look” (Erdelyi 1974, p. 2).The last two decades have seen the pendulum swing

again, away from a robust division between perceptualand cognitive processing and back toward the previouslyfashionable New Look understanding of perception. Thedriving force in recent years has been a tidal wave ofstudies seeming to show influences on perception fromall corners of the mind. However, the particular theoreticalmotivations behind these various results are nonuniform, soit will be useful to understand these studies in groups.Roughly, today’s alleged top-down effects on perceptionare effects of motivation, action, emotion, categorization,and language.

3.1. Motivation

Those recent results with the greatest overlap with the NewLook movement concern influences of motivation (desires,needs, values, etc.) on perception. For example, it hasrecently been reported that desirable objects such as choc-olate look closer than undesirable objects such as feces(Balcetis & Dunning 2010; see also Krpan & Schnall2014); that rewarding subjects for seeing certain interpreta-tions of ambiguous visual stimuli actually makes the stimulilook that way (Balcetis & Dunning 2006; see also Pascucci& Turatto 2013); that desirable destinations seem closerthan undesirable ones (Alter & Balcetis 2011; see also Bal-cetis et al. 2012); and even that women’s breasts appearlarger to sex-primed men (den Daas et al. 2013). Otherstudies have focused on physiological needs. Forexample, muffins are judged as larger by dieting subjects(van Koningsbruggen et al. 2011), food-related words areeasier to identify when observers are fasting (Radel &Clément-Guillotin 2012), and ambiguous surfaces arejudged as more transparent (or “water-like”) by subjectswho eat salty pretzels and become thirsty (Changizi &Hall 2001; Fig. 2B). Morally relevant words reportedly“pop out” in visual awareness when briefly presented(Gantman & Van Bavel 2014; Fig. 2C), and follow-upstudies suggest that the effect may arise from a desire forjustice. Many of these contemporary studies explicitly

take inspiration from the New Look, claiming to studythe same phenomena but “armed with improved methodo-logical tools and theories” (Dunning & Balcetis 2013,p. 33).

3.2. Action

Another class of recent top-down effects concerns action-based influences on perception. Physical burdens thatmake actions more difficult reportedly make the environ-ment look more imposing: wearing a heavy backpackinflates estimates of distance (Proffitt et al. 2003), as doesthrowing a heavy ball (Witt et al. 2004); fatigued or unfitindividuals overestimate slant and distance relative torested or fit individuals (Bhalla & Proffitt 1999; Coleet al. 2013; Sugovic & Witt 2013; Fig. 2D); fixing weightsto subjects’ ankles increases size estimates of jumpablegaps (Lessard et al. 2009); holding one’s arms out decreaseswidth estimates of doorway-like apertures (Stefanucci &Geuss 2009; Fig. 2E); and standing on a wobbly balancingboard reduces width estimates of a walkable beam (Geusset al. 2010). Conversely, improvements in ability arereported to shrink the perceived environment to makeactions look easier: Subjects who hold reach-extendingbatons judge targets to be closer (Witt et al. 2005; seealso Abrams & Weidler 2015); subjects who drink asugary beverage (rather than a low-calorie alternative) esti-mate hills as shallower (Schnall et al. 2010); and swimmerswho wear flippers judge underwater targets as closer (Wittet al. 2011). Similarly, exceptional athletic performance isreported to alter the perceived size of various types ofsporting equipment, yielding perceptual reports of largersoftballs (Gray 2013; Witt & Proffitt 2005), wider footballgoal posts (Witt & Dorsch 2009), lower tennis nets (Witt& Sugovic 2010), larger dartboards (Cañal-Bruland et al.2010; Wesp et al. 2004; Fig. 2F), larger golf holes (Wittet al. 2008), and (for parkour experts) shorter walls(Taylor et al. 2011). This approach emphasizes theprimacy of action in perception (inspired in many ways byGibson 1979), holding that action capabilities directlyalter the perceived environment (for reviews, see Proffitt2006; Proffitt & Linkenauger 2013; Witt 2011a). (Thoughit is not entirely clear whether action per se is a truly cog-nitive process, we mean to defend an extremely broadthesis regarding the sorts of states that cannot affect per-ception, and this most definitely includes action. Moreover,in many of these cases, it has been proposed that it is notthe action that penetrates perception but rather the inten-tion to act – e.g., Witt et al. 2005 – in which case sucheffects would count as alleged examples of cognition affect-ing perception after all.)

3.3. Affect and emotion

A third broad category of recently reported top-downeffects involves affective and emotional states. In suchcases, the perceived environment is purportedly alteredto match the perceiver’s mood or feelings. For example,recent studies report that thinking negative thoughtsmakes the world look darker (Banerjee et al. 2012; Meieret al. 2007; Fig. 2G); fear and negative arousal make hillslook steeper, heights look higher, and objects look closer(Cole et al. 2012; Harber et al. 2011; Riener et al. 2011;Stefanucci & Proffitt 2009; Stefanucci & Storbeck 2009;




Stefanucci et al. 2012; Storbeck & Stefanucci 2014; Teach-man et al. 2008); scary music makes ambiguous images(e.g., an ambiguous figure that might be an alligator or asquirrel) take on their scarier interpretations (Prinz &Seidel 2012; Fig. 2H); social exclusion makes otherpeople look closer (Pitts et al. 2014); and smiling facesappear brighter (Song et al. 2012; Fig. 2I). Here, theeffects are either thought to accentuate one’s emotionalstate – perhaps because affect is informative about theorganism’s needs (e.g., Storbeck & Clore 2008) – or toenergize the perceiver to counteract such negative feelings.

3.4. Categorization and language

A final class of contemporary top-down effects concernscategories and linguistic labels. A popular testing groundfor such effects has involved the perception of color andlightness. For example, it has been reported that learningcolor–letter associations biases perceptual judgmentstoward the learned hues (Goldstone 1995; Fig. 2J); catego-rizing faces as Black or White alters the faces’ perceivedskin tones, even when the faces are in fact equally luminant(Levin & Banaji 2006); and knowledge of an object’s typicalcolor (e.g., that bananas are yellow) makes grayscale imagesof those objects appear tinged with their typical colors(Hansen et al. 2006; Witzel et al. 2011; e.g., Fig. 2K). Con-ceptual categorization is also reported to modulate variousvisual phenomena. For example, the Ebbinghaus illusion,in which a central image appears smaller when surroundedby large images (or larger when surrounded by smallimages), is reportedly stronger when the surroundingimages belong to the same conceptual category as thecentral image (Fig. 2L; Coren & Enns 1993; see also vanUlzen et al. 2008).

Similar effects may arise from linguistic categories andlabels. For example, the use of particular color terms isreported to affect how colors actually appear (e.g.,Webster & Kay 2012), and labeling visual stimuli reportedlyenhances processing of such stimuli and may even altertheir appearance (Lupyan & Spivey 2008; Lupyan et al.2010; Lupyan & Ward 2013; Fig. 2M). Other alleged lin-guistic effects include reports of visual motion aftereffectsafter reading motion-related language (e.g., “Google’sstock sinks lower than ever”; Dils & Boroditsky 2010a;2010b; see also Meteyard et al. 2007), and differences inthe apparent motion of a Chinese character’s strokedepending on knowledge of how such characters arewritten (Tse & Cavanagh 2000; though see Li & Yeh2003; Fig. 2N).

Note that the effects cited in this section are not onlynumerous and varied, but also they are exceptionallyrecent: Indeed, the median publication year for theempirical papers cited in section 3 is 2010.

4. The six “pitfalls” of top-down effects onperception

If there are no top-down effects of cognition on perception,then how have so many studies seemed to find such richand varied evidence for them? A primary purpose of thispaper is to account for the wealth of research reportingsuch top-down effects. We suggest that this research falls

prey to a set of “pitfalls” that undermine their claims.These pitfalls have four primary features:

1. They are few in number.We suggest that nearly all ofthe recent literature on top-down effects is susceptible to asurprisingly small group of such pitfalls.2. They are empirically anchored. These are not idle

suspicions about potential causes of such effects, butrather they are empirically grounded – not just in theweak sense that they discuss relevant empirical evidence,but in the stronger sense that they have demonstrablyexplained several of the most prominent apparent top-down effects on perception, in practice.3. They are general in scope. Beyond our concrete case

studies, we also aim to show that the pitfalls are broadlyapplicable, with each covering dozens more top-downeffects.4. They are theoretically rich. Exploring these pitfalls

raises several foundational questions not just about percep-tion and cognition, but also about their relationships withother mental processes, including memory, attention, andjudgment.

We contend that any apparent top-down effect that fallsprey to one or more of these pitfalls would be compro-mised, in the sense that it could be explained by deflation-ary, routine, and certainly nonrevolutionary factors. It isthus our goal to establish the empirical concreteness andgeneral applicability of these pitfalls, so that it is clearwhere the burden of proof lies: No claim of a top-downeffect on perception can be accepted until these pitfallshave been addressed.We first discuss each pitfall in general terms and then

provide empirical case studies of how it can be exploredin practice, along with suggestions of other top-downeffects to which it may apply. In each case, we concludewith concrete lessons for future research.

4.1. Pitfall 1: An overly confirmatory research strategy

In general, experimental hypotheses can be tested in twosorts of ways: Not only should you observe an effectwhen your theory calls for it, but also you should notobserve an effect when your theory demands itsabsence. Although both kinds of evidence can be deci-sive, the vast majority of reported top-down effects onperception involve only the first sort of test: a hypothesisis proffered that some higher-level state affects what wesee, and then such an effect is observed. Though it isperhaps unsurprising that these studies only test such“confirmatory predictions,” in our view this strategyessentially misses out on half of the possible decisive evi-dence. Recently, this theoretical perspective has beenmade empirically concrete by studies testing certainkinds of uniquely disconfirmatory predictions ofvarious top-down phenomena.

4.1.1. Case studies. To make the contrast between confir-matory and disconfirmatory predictions concrete, we con-ducted a series of studies (Firestone & Scholl 2014b)inspired by an infamous art-historical reasoning errorknown as the “El Greco fallacy.” Beyond appreciating thevirtuosity of his work, the art-history community has longpuzzled over the oddly elongated human figures in ElGreco’s paintings. To explain these distortions, it was




once supposed that El Greco suffered from an uncom-monly severe astigmatism that effectively “stretched” hisperceived environment, such that El Greco had simplybeen painting what he saw. This perspective was oncetaken seriously, but upon reflection it involves a conceptualconfusion: If El Greco had truly experienced a stretched-out world, then he would also have experienced anequally stretched-out canvas, canceling out the supposedreal-world distortions and thus leaving no trace of themin his reproductions. The distortions in El Greco’s paint-ings, then, could not reflect literal perceptual distortions(Anstis 2002; Firestone 2013b).We exploited the El Greco fallacy to show that multiple

alleged top-down effects cannot genuinely be effects onperception. Consider, for example, the report that reflect-ing on unethical actions makes the world look darker(Banerjee et al. 2012; Fig. 2O). The original effect wasobtained using a numerical scale: After reflecting onethical or unethical actions, subjects picked a number onthe scale to rate the brightness of the room they werein. We replicated this effect with one small change:Instead of a numerical scale, subjects used a scale ofactual grayscale patches to rate the room’s brightness.According to the view that reflecting on negative actionsmakes the world look darker, this small change drasticallyalters the study’s prediction: If the world really looksdarker, then the patches making up the scale shouldlook darker too, and the effects should thus cancel eachother out (just as the alleged distortions in El Greco’sexperience of the world would be canceled out by hisequally distorted experience of his canvas). However, thefollow-up study succeeded: subjects still picked a darkerpatch to match the room after reflecting on an unethicalaction (Firestone & Scholl 2014b, Experiment 5). Thiseffect, then – like the distortions in El Greco’s work –must not reflect the way the world actually looked tosubjects.This approach is in no way limited to the particulars of

the morality/brightness study. Indeed, to apply the samelogic more broadly, we also explored a report of a very dif-ferent higher-level state (a subject’s ability to act in acertain way) on a very different visual property (perceiveddistance). In particular, holding a wide rod across one’sbody (Fig. 2E) reportedly makes the distance betweentwo poles (which form a doorway-like aperture) look nar-rower, as measured by having subjects instruct the experi-menter to adjust a measuring tape to visually match theaperture’s width. The effect supposedly arises becauseholding the rod makes apertures less passable (Stefanucci& Geuss 2009). We successfully replicated this result, butwe also tested it with one critical difference: Instead ofadjusting a measuring tape to record subjects’ width esti-mates, the experimenter used two poles that themselvesformed an independent and potentially passable aperture.Again, the El Greco logic applies: If holding a rod reallydoes perceptually compress apertures, then this variantshould “fail,” because subjects should see both aperturesas narrower. But the experiment did not “fail”: Subjectsagain reported narrower apertures even when respondingwith an aperture (Firestone & Scholl 2014b, Experiment2). Therefore, this effect cannot reflect a true perceptualdistortion – not because the effect fails to occur, butrather because it occurs even when it shouldn’t. (In laterexperiments, we determined the true, nonperceptual,

explanation for this effect, involving task demands; seePitfall 3.)

4.1.2. Other susceptible studies. As an example of testingdisconfirmatory predictions, the El Greco fallacy applies toany constant-error distortion that should affect equally themeans of reproduction (e.g., canvases, grayscale patches)and the item reproduced (e.g., visual scenes to bepainted, bright rooms). The studies that fail to test suchpredictions are too numerous to count; essentially, nearlyevery study falls into this category. However, somestudies of top-down effects on perception may havetested such predictions inadvertently – and, given theirresults, perhaps committed the El Greco fallacy.Consider, for example, the report that after repeatedly

viewing one set of red and violet letters and a second setof blue and violet numbers, subjects judged token violetletters to look redder than they truly were and tokenviolet numbers to look bluer than they truly were (Gold-stone 1995; Fig. 2J). This effect was measured by havingsubjects adjust the hue of a stimulus to perceptuallymatch the symbol being tested. However, the adjustedstimulus was a copy of that symbol! For example, afterviewing a red “T,” a reddish-violet “L,” and a violet “E,”subjects judged the E to be redder – as measured by adjust-ing the hue of a second E. This commits the El Grecofallacy: if Es really look redder after one sees other redletters, then both the to-be-matched E and the adjustableE should have looked redder, and the effects should havecanceled one another out. That such an effect was never-theless obtained suggests it cannot be perceptual.Similarly, consider the following pair of results, reported

together: Subjects judged gray patches to be darker afterreading negative (vs. positive) words, and subjects judgedwords printed in gray ink to be darker if the words werenegative (vs. positive), as measured by selecting a darkergrayscale patch to match the word’s lightness (Meieret al. 2007; Fig. 2G). Here too is an El Greco fallacy: if,per the first result, reading negative words makes graypatches look darker, then the gray patches from thesecond result should also have looked darker, and theeffects of one should have canceled out the other.The El Greco fallacy may also afflict reports that linguis-

tic color categories alter color appearance (Webster & Kay2012). For example, a color that is objectively between blueand green may appear either blue or green because ourcolor terms single out those colors when they discretizecolor space, creating clusters of perceptual similarity.However, such studies use color spaces specifically con-structed for perceptual uniformity, such that each stepthrough the space’s parameters is perceived as equal inmagnitude. This raises a puzzle: If color terms affect per-ceived color, then such effects should already have beenassimilated into the color space, leaving no room for colorterms to exert their influence in studies using such colorspaces. That these studies still show labeling effects sug-gests an alternative explanation.2

4.1.3. A lesson for future research. To best determine theextent to which cognition influences perception, futurestudies should proactively employ both confirmatory anddisconfirmatory research strategies; to do otherwise is toignore half of the predictions the relevant theories gener-ate. In pursuing disconfirmatory evidence, El Greco–




inspired research strategies in particular have three distinctadvantages. First, they can rule out perceptual explanationswithout relying on null effects and their attendant interpre-tive problems; instead, this strategy can disconfirm top-down interpretations through positive replications.Second, the El Greco strategy can fuel such implicationseven before researchers determine the actual (nonpercep-tual) culprit (just as we know that astigmatism does notexplain El Greco’s distortions, even if we remain uncertainwhat does explain them). Finally, this strategy is broadlyrelevant – being applicable any time a scale can be influ-enced just as the critical stimuli are supposedly influenced(e.g., in nearly all perceptual matching tasks).

4.2. Pitfall 2: Perception versus judgment

Many alleged top-down effects on perception live near theborder of perception and cognition, where it is not alwaysobvious whether a given cognitive state affects what wesee or instead only our inferences or judgments made onthe basis of what we see. This distinction is intuitive else-where. For example, whereas we can perceive the coloror size of some object – say, a shoe –we can only infer orjudge that the object is expensive, comfortable, or fashion-able (even if we do so based on how it looks). Top-downeffects on perception pose a special interpretive challengealong these lines, especially when they rely on subjects’verbal reports. Whereas expensiveness can only bejudged (not perceived), other properties such as colorand size can be both perceived and judged: We can directlysee that an object is red, and we can also conclude or inferthat an object is red. For this reason, any time an experi-ment shifts perceptual reports, it is possible that the shiftreflects changes in judgment rather than perception. Andwhereas top-down effects on perception would be revolu-tionary and consequential, many top-down effects on judg-ments are routine and unsurprising, carrying fewimplications for the organization of the mind. (Of course,that is not to say that research on judgment in general isnot often of great interest and import – just that someeffects on judgment truly are routine and universallyaccepted, and those are the ones that may explain awaycertain purported top-down effects of cognition onperception.)

Though the distinction between perception and judg-ment is often clear and intuitive – in part because theycan so clearly conflict (as in visual illusions) –we contendthat judgment-based alternative explanations for top-down effects have been severely underappreciated inrecent work. Fortunately, there are straightforwardapproaches for teasing them apart.

4.2.1. Case studies. It has been reported that throwing aheavy ball (rather than a light ball) at a target increases esti-mates of that target’s distance (Witt et al. 2004). One inter-pretation of this result (favored by the original authors) isthat the increased throwing effort actually made thetarget look farther away, and that this is why subjectsgave greater distance estimates. However, another possibil-ity is that subjects only judged the target to be farther, evenwithout a real change in perception. For example, afterhaving such difficulty reaching the target with theirthrows, subjects may have simply concluded that thetarget must have been farther away than it looked.

A follow-up study tested these varying explanations anddecided the issue in favor of an effect on judgment ratherthan perception. Whereas the original study asked for esti-mates of distance without specifying precisely how subjectsshould make such estimates, Woods et al. (2009) systemati-cally varied the distance-estimation instructions, contrast-ing cases (between subjects) asking for reports of how farthe target “visually appears” with cases asking for reportsof “how far away you feel the object is, taking all nonvisualfactors into account” (p. 1113). In this last condition, sub-jects were especially encouraged to separate perceptionfrom judgment: “If you think that the object appears tothe eye to be at a different distance than it feels (takingnonvisual factors into account), just base your answer onwhere you feel the object is.” Tellingly, the effect ofeffort on distance estimation replicated only in the “nonvi-sual factors” group, and not in the “visually appears”group – suggesting that the original results reflected whatsubjects thought about the distance rather than how thedistance truly looked to them.Similarly, it was reported that accuracy in throwing darts

at a target affected subsequent size judgments of the target,which was initially assumed to reflect a perceptual change:Less-accurate throwing led to smaller target-size estimates,as if one’s performance perceptually resized the target(Wesp et al. 2004; Fig. 2F). However, the same researchersrightly wondered whether this was genuinely an effect onperception or whether these biased size estimates mightinstead be driven by overt inferences that the target musthave been smaller than it looked (perhaps to explain orjustify poor throwing performance). To test this alternative,the same research group (Wesp & Gasper 2012) replicatedthe earlier result – but then ran a follow-up condition inwhich, before throwing, subjects were told that the dartswere faulty and inaccurate. This additional instruction elim-inated the correlation between performance and reportedsize. With a ready-made explanation already in place, sub-jects no longer needed to “blame the target”: Rather thanconclude that their poor throwing resulted from a smalltarget, subjects instead attributed their performance tothe supposedly faulty darts and thus based their size esti-mates directly on how the target looked.Note that this is a perfect example of the kind of judg-

ment that can only be described as “routine.” Even ifother sorts of top-down effects on judgment more richlyinteract with foundational issues in perception research,blaming a target for one’s poor performance is not one ofthem.

4.2.2. Other susceptible studies. Many alleged top-downeffects on perception seem explicable by appeal to thesesorts of routine judgments. One especially telling patternof results is that many of these effects are found evenwhen no visual stimuli are used at all. For example,factors such as value and ease of action have beenclaimed to affect online distance perception (e.g., Balcetis& Dunning 2010; Witt et al. 2005), but those samefactors have been shown to affect the estimated distanceof completely unseen (and sometimes merely imagined)locations such as Coney Island (Alter & Balcetis 2011) orone’s work office (Wakslak & Kim 2015). Clearly sucheffects must reflect judgment and not perception – yettheir resemblance to other cases that are indeed claimed




as top-down effects on perception suggests that many suchcases could reflect judgmental processes after all.Other cases seem interpretable along these lines all on

their own. For example, another study also demonstratedan effect of dart-throwing performance on size judgments –but found that the effect disappeared when subjects madetheir throws while hanging onto a rock-climbing wall 12feet above the ground (Cañal-Bruland et al. 2010).Though this phenomenon was interpreted as an effect ofanxiety on action-specific perception, the finding couldeasily be recast as an effect on judgment instead: Subjectswho performed poorly while clinging to the rock-climbingwall had an obvious explanation for performing poorlyand so had no need to explain their misses by inflatingtarget-size estimates.In other cases, the inference to judgment rather than

perception can be more straightforward. For example,politically conservative subjects rated darkened images ofBarack Obama as more “representative” of him than light-ened images, whereas liberal subjects showed the oppositepattern (Caruso et al. 2009), and this effect was interpretedas an effect of partisan attitudes on perceived skin tone.However, it seems more likely that darker photos (ordarker skin tones) seemed more negative to subjects, andthat conservatives deemed them more representative(and liberals less representative) for that reason – becauseconservatives think more negatively about Obama than lib-erals do. (By analogy, we suspect that conservative subjectswould also rate a doctored image of Obama with bright redhorns on his forehead as more “representative” than animage of Obama with a halo, and that liberals wouldshow the opposite pattern; but clearly such a result wouldnot imply that conservatives literally see Obama as havinghorns!) Other purported top-down effects that seem simi-larly explicable include effects on visually estimatedweight (Doerrfeld et al. 2012), the estimated looming ofspiders (Riskind et al. 1995), and the rated anger inAfrican-American or Arab faces (Maner et al. 2005).

4.2.3. A lesson for future research. The distinctionbetween perception and judgment is intuitive and uncon-troversial in principle, but it is striking just how few discus-sions of top-down effects on perception even mentionjudgmental effects as possible alternative explanations.(For some exceptions see Alter & Balcetis 2011; Lupyanet al. 2010; Witt et al. 2010.) Future work relying on subjec-tive perceptual reports must attempt to disentangle thesepossibilities. It would of course be preferable for suchstudies to empirically distinguish perception from judg-ment – for example, by using performance-based measuresin which subjects’ success is tied directly to how they per-ceive the stimuli (such as a visual search task; cf. Scholl &Gao 2013). Or, per the initial case study reviewed above,future work can at least ask the key questions in multipleways that differentially load on judgment and perception.At a minimum, given the importance of distinguishing

judgment from perception, it seems incumbent on any pro-posal of a top-down effect to explicitly and prominentlyaddress the distinction, even if only rhetorically – becausea shift from perception to judgment may dramaticallyreduce such an effect’s potential revolutionary conse-quences. And at the same time, we note that certainterms may actively obscure this issue and so should beavoided. For example, many papers in this literature

advert to effects on “perceptual judgment” (e.g., Meieret al. 2007; Song et al. 2012; Storbeck & Stefanucci2014), which can only invite confusion about this founda-tional distinction.

4.3. Pitfall 3: Demand and response bias

Vision experiments occur in a variety of controlled environ-ments (including the laboratory), but any such environmentis also inevitably a social environment –which raises thepossibility that social biases may intrude on perceptualreports in a more specific way than we saw in Pitfall2. Whereas judgments of various visual qualities are oftensincerely held even when they are subject to top-downinfluence (such that, e.g., inaccurate dart-throwers maytruly believe that the target must be smaller than itlooks), other sorts of biases may reflect more active modu-lation of responses by participants – such that this pitfall isconceptually distinct from the previous one. In particular,the social nature of psychology experiments can readilylead to reports (of anything, including percepts) being con-taminated by task demands, wherein certain features ofexperiments lead subjects to adjust their responses (eitherconsciously or unconsciously) in accordance with theirassumptions about the experiment’s purpose (or the exper-imenters’ desires). (For a review of the power and perva-siveness of such effects, see Rosenthal & Rubin 1978.)Contamination by demand characteristics seems espe-

cially likely in experiments involving a single conspicuousmanipulation and a single perceptual report. But evenmore so than with the previous pitfall, it seems especiallyeasy to combat such influences – for example, by askingsubjects directly about the experiment and/or by directlymanipulating their expectations.

4.3.1. Case studies. Consider the effect of wearing a heavybackpack on slant estimates (Bhalla & Proffitt 1999;Fig. 2D). One possibility is that backpacks make hills looksteeper, and that the subjects faithfully reported whatthey saw. But another explanation is that subjects modifiedtheir responses to suit the experimental circumstances, inwhich a very conspicuous manipulation (a curiously unex-plained backpack) was administered before obtaining asingle perceptual judgment (regarding the hill’s slant).A recent series of studies shows that the experimental

demand of wearing a backpack can completely accountfor the backpack’s effect on slant estimates. When back-pack-wearing subjects were given a compelling (but false)cover story to justify the backpack’s purpose (to holdheavy monitoring equipment during climbing), the effectof heavy backpacks on slant estimation completely disap-peared (Durgin et al. 2009; see also Durgin et al. 2012).With a plausible cover story, subjects had very differentexpectations about the experiment’s purpose (expectationsthat they articulated explicitly during debriefing), which nolonger suggested that the backpack “should” modulatetheir responses. Similar explanations have subsequentlybeen confirmed for other effects of action on perceptualreports, including effects of aperture “passability” onspatial perception (Firestone & Scholl 2014b) and energyon slant perception (Durgin et al. 2012; Shaffer et al.2013). For example, no effect of required climbing effortis found without a transparent manipulation – such aswhen subjects estimate the slant of either an (effortful)




staircase or an (effort-free) escalator in a between-subjectsdesign (Shaffer & Flint 2011).

Other studies have implicated task demands in very dif-ferent top-down effects. For example, it has been reportedthat, when subjects can win a gift card taped to the groundif they throw a beanbag closer to the gift card than theirpeers do, subjects undershoot the gift card if it is worth$25 but not if it is worth $0 – suggesting (to the originalauthors) that more desirable objects look closer (Balcetis& Dunning 2010; Fig. 2P). However, in addition to thevalue of the gift card, the demands of the task differedacross these conditions in an especially intuitive way:Whereas subjects may employ various throwing strategiesin earnest attempts to win a $25 gift card, they may nottry to “win” a $0 gift card (which is a decidedly odd task).For example, subjects who are genuinely trying to winthe $25 gift card might undershoot the card if they believedit would be awarded to the closest throw without goingover, or if they anticipated that the beanbag wouldbounce closer to the gift card after its first landing.However, they may not show those biases for the $0 giftcard, which wouldn’t have been worth any such strategiz-ing. A follow-up study (Durgin et al. 2011a) tested thesepossibilities directly and found that slightly changing theinstructions so that the challenge was to hit the gift carddirectly (rather than land closest) led subjects to throwthe beanbag farther (perhaps because they were nolonger worried that it would bounce or that they wouldbe disqualified if they overshot), just as would be expectedif differences in strategic throwing (rather than differencesin actual perception) explained the initial results.

4.3.2. Other susceptible studies. Perhaps no pitfall is asgenerally applicable as demand and response bias, espe-cially for studies relying entirely on observer reports. Agreat many reported top-down effects on perception usevery salient manipulations and ask for perceptual judg-ments that either give subjects ample opportunity to con-sider the manipulation’s purpose or make the “right”answer clear. For example, it has been reported that,when shown a range of yellow-orange discs superimposedon a traffic light’s middle bulb, German subjects (forwhom that light’s color is called gelb, or yellow) classifiedmore discs as “yellow” than did Dutch subjects (who callit oranje, or orange; Mitterer et al. 2009; Fig. 2Q).Though interpreted as an effect of language on percep-tion – the claim being that the German subjects visuallyexperienced the colored discs as yellower – it seems justas plausible that the subjects were simply following conven-tion, assigning the yellow-orange discs the socially appro-priate names for that context.

Many other studies use salient manipulations and mea-sures in a manner similar to the backpacks and hills exper-iments (Bhalla & Proffitt 1999). For example, similarexplanations seem eminently plausible for reportedeffects of desirability on distance perception (e.g., the esti-mated distance of feces vs. chocolate; Balcetis & Dunning2010), of racial identity on faces’ perceived lightness (Levin& Banaji 2006), of stereotypes on the identity of weaponsand tools (Correll et al. 2015), of tool use on the perceiveddistance to reachable targets (Witt et al. 2005), of scarymusic on the interpretation of scary or nonscary ambiguousfigures (Prinz & Seidel 2012; Fig. 2H), and of fear of

heights on perceived height (Clerkin et al. 2009; Stefanucci& Proffitt 2009).

4.3.3. A lesson for future research. In light of recent find-ings concerning task demands in studies of top-downeffects on perception (especially Durgin et al. 2009), it isno longer possible to provide compelling evidence for atop-down effect on perception without considering theexperiment’s social context. Yet, so many studies nevereven mention the possibility of demand-based effects(including several studies mentioned above, e.g., Mittereret al. 2009; Prinz & Seidel 2012). (For some exceptions,see Levin & Banaji 2006; Schnall et al. 2010; Witt2011b.) This is especially frustrating because assessingdemand effects is often easy and cost-free. In particular,although demand effects can be mitigated by nontranspar-ent manipulations or indirect measures, they can also oftenbe assessed by simply asking the subjects about the experi-ment – for example, during a careful postexperimentdebriefing. For example, before the experiment’s purposeis revealed, researchers can carefully ask subjects whatthey thought the experiment was about, what strategiesthey used, and so forth. These sorts of questions canreadily reveal (or help rule out) active demand factors.Such debriefing was especially helpful, for example, in

the case of backpacks and reported slant, wherein manysubjects explicitly articulated the experimental hypothesiswhen asked – and only those subjects showed the backpackeffect (Durgin et al. 2009). In this way, we believe Durginet al.’s 2009 report has effectively set the standard for suchexperiments: Given the negligible costs and the potentialintellectual value of such careful debriefing, we contendthat claims of top-down effects (especially in studies usingtransparent manipulations) can no longer be crediblewithout at least asking about – and reporting – subjects’beliefs about the experiment.

4.4. Pitfall 4: Low-level differences (and amazingdemonstrations!)

Whereas many studies search for top-down effects on per-ception by manipulating states of the perceiver (e.g., moti-vations, action capabilities, or knowledge), many other top-down effects involve manipulations of the stimuli usedacross experimental conditions. For example, one way totest whether arousal influences spatial perception couldbe to test a high-arousal group and a low-arousal groupon perception of the same stimulus (e.g., a precariousheight; Teachman et al. 2008). However, another strategycould be to measure how subjects perceive the distanceof arousing versus nonarousing stimuli (e.g., live tarantulasvs. plush toys; Harber et al. 2011). Though both approacheshave strengths and weaknesses, one difficulty in manipulat-ing stimuli across experimental conditions is the possibilitythat the intended top-down manipulation (e.g., the evokedarousal) is confounded with changes in the low-level visualfeatures of the stimuli (e.g., as live tarantulas might differ insize, color, and motion from plush toys) – and that thoselow-level differences might actually be responsible for per-ceptual differences across conditions.We have suggested (and will continue to suggest) that

many of the pitfalls we discuss here have been largelyneglected by the literature on top-down effects, but thispitfall is an exception: Studies that manipulate stimuli




often do acknowledge the possibility of low-level differ-ences (and on occasion actively attempt to control forthem). Nevertheless, we contend that such low-level differ-ences are even more pervasive and problematic than hasbeen realized, and that simple experimental designs canreveal when such differences are responsible for apparenttop-down effects.

4.4.1. Case studies. One especially compelling and cur-rently influential top-down effect on perception is areport that Black (i.e., African-American) faces lookdarker than White (i.e., Caucasian) faces, even whenmatched for mean luminance, as in Figure 3A (Levin &Banaji 2006). This finding is today widely regarded as oneof the strongest counterexamples to modularity (e.g.,Collins & Olson 2014; Macpherson 2012; Vetter &Newen 2014) – no doubt because, in addition to thecareful experiments reported in the paper, the differencein lightness is clearly apparent upon looking at thestimuli. In other words, this top-down effect works as a“demonstration” as well as an experiment.That last point is worth emphasizing given the preva-

lence of “demos” in vision science. In our field, experimen-tal data about what we see are routinely accompanied bysuch demonstrations – in which interested observers canexperience the relevant phenomena for themselves, oftenin dramatic fashion. For example, no experiments areneeded to convince us of the existence of phenomenasuch as motion-induced blindness, apparent motion, orthe lightness illusions depicted in Figure 1. Of course,that is not to say that demos are necessary for progress invision science; most experiments surely get by withoutthem. But effective demos can provide especially compel-ling evidence, and they may often directly rule out thekinds of worries we expressed in discussing the previoustwo pitfalls (i.e., task demands and postperceptualjudgments).In this context, the demonstration of race-based light-

ness distortions (Levin & Banaji 2006) is exceptional,insofar as it is one of the only such demos in this large lit-erature. Indeed, it strikes us as an awkward fact that sofew such effects can actually be experienced for oneself.For example, the possibility that valuable items lookcloser is testable not only in a laboratory (e.g., Balcetis &Dunning 2010) but also from the comfort of home: Rightnow you can place a $20 bill next to a $1 bill and see foryourself whether there is a perceptual difference. Similarly,knowledge of an object’s typical color (e.g., that bananas areyellow) reportedly influences that object’s perceived color,

such that a grayscale image of a banana is judged to bemore than 20% yellow (Hansen et al. 2006; Olkkonenet al. 2008); however, if you look now at a grayscaleimage of a banana (Fig. 2K), we predict that you will notexperience this effect for yourself – even though thereported effect magnitudes far exceed established discrim-ination thresholds (e.g., Hansen et al. 2008; Krauskopf &Gegenfurtner 1992). (You may notice that many of thetop-down effects in Figure 2 are caricatured, e.g., withactual luminance differences for positive vs. negativewords and smiling vs. frowning faces. This is becausewhen the effects weren’t caricatured in this way, readerscould not understand the claims – because they could notexperience the effect!)All of this makes the reported lightness difference much

more compelling: As you may experience in Figure 3A, theBlack face truly looks darker than the luminance-matchedWhite face. But is this a top-down effect on perception?Though the face stimuli were matched for mean lumi-nance, there are of course many visual cues to lightnessthat are independent of mean luminance. For example,in many lightness illusions, two regions of equal luminancenevertheless appear to have different lightnesses becauseof depicted patterns of illumination and shadow (as inFig. 1). Indeed, a close examination of the face stimuli inFigure 3A suggests that the Black face seems to be underillumination, whereas the White face doesn’t look particu-larly illuminated or shiny – a difference that has long beenknown to influence perceived lightness (Adelson 2000; Gil-christ & Jacobsen 1984). And the Black face has a darkerjawline, whereas the White face has darker eyes. Ofcourse, there must exist some low-level differencesbetween the images, because otherwise they would beidentical; nevertheless, the question remains whethersuch lower-level visual factors are responsible for theeffect, rather than the meaning or category (here, race)that is correlated with that low-level difference.To test whether one or more such low-level differences –

rather than race, per se – explain the difference in perceivedlightness, we replicated this study with blurred versions ofthe face stimuli, so as to eliminate race information whilepreserving many low-level differences in the images (includ-ing the match in average luminance and contrast) – as inFigure 3B (Firestone & Scholl 2015a). After blurring, thevast majority of observers asserted that the two faces actuallyhad the same race (or were even the same person).However, even those observers who asserted that thefaces had the same race nevertheless judged the blurryimage derived from the Black face to be darker than theblurry image derived from the White face. This result

(a) (b)

Figure 3. (A) The face stimuli (matched in mean luminance) from Experiment 1 of Levin and Banaji (2006). (B) The blurred versions ofthe same stimuli used in Firestone and Scholl (2015a), which preserved the match in mean luminance but obscured the race information.




effectively shows how the lightness difference can derivefrom low-level visual features –which, critically, arepresent in the original images –without any contributionfrom perceived race. (And note that such results are unlikelyto reflect unconscious race categorization; it would be a dis-tinctively odd implicit race judgment that could influenceexplicit lightness judgments but not explicit race judgments.)And although the original effect (with unblurred faces)could of course still be explained entirely by race (ratherthan by the lower-level differences now shown to affect per-ceived lightness), it is clear that further experiments wouldbe needed to show this – and so we conclude that theinitial demonstration of Levin and Banaji (2006) providesno evidence for a top-down effect on perception.3

Many other effects that initially seemed to reflect high-level factors have been shown to reflect lower-level visualdifferences across conditions. Consider, for example,reports that categorical differences facilitate visual searchwith line drawings of animals and artifacts – e.g., withfaster and more efficient searches for animals among arti-facts, and vice versa (Levin et al. 2001). This result initiallyappeared to be a high-level effect on a fairly low-level per-ceptual process, given that efficient visual search is typicallyconsidered “preattentive.” However, on closer investiga-tion (and to their immense credit), the same researchersdiscovered systematic low-level differences in theirstimuli –wherein the animals (e.g., snakes, fish) had morecurvature than the artifacts (e.g., chairs, briefcases) –which sufficiently explained the search benefits (as revealedby follow-up experiments directly exploring curvature).

4.4.2. Other susceptible studies. The possibility of suchlow-level confounds is a potential issue, almost by defini-tion, with any top-down effect that varies stimuli acrossconditions. For example, size contrast is reportedlyenhanced when the inducing images are conceptuallysimilar to the target image – such that a central dogimage looks smaller when surrounded by images of largerdogs versus images of larger shoes (Coren & Enns 1993).However, dogs are also more geometrically similar toeach other than they are to shoes (for example, the shoeimages were shorter, wider, and differently oriented thanthe dog images; Fig. 2L), and size contrast may insteadbe influenced by such geometric similarity. (Coren andEnns are quite sensitive to this concern, but their follow-up experiments still involve important geometric differ-ences of this sort.)

Other investigations of top-down effects on size contrastalso manipulate low-level properties, for example contrast-ing natural scenes with different distributions of color andcomplexity (e.g., van Ulzen et al. 2008). Or, in a very differ-ent case, studies of how fear may affect spatial perceptionoften involve stimuli with very different properties (e.g., alive tarantula vs. a plush toy; Harber et al. 2011), or eventhe same stimulus viewed from very different perspectives(e.g., a precarious balcony viewed from above or below;Stefanucci & Proffitt 2009).

4.4.3. A lesson for future research. Manipulating theactual stimuli or viewing circumstances across experimentalconditions is a perfectly viable methodological choice, but itadds a heavy burden to avoid low-level differences betweenthe stimuli. Critically, this burden can be met in at least twoways. One possibility is to preserve the high-level factor

while eliminating the low-level factor. (In other contextslooking at fearful stimuli, for example, images of spidershave been contrasted not with plush toys, but with “scram-bled” spider images, or even images of the same line seg-ments rearranged into a flower; e.g., New & German2015.) Another possibility – as in our study of race catego-ries and lightness (Firestone & Scholl 2015a) – is to pre-serve the low-level factor while eliminating the high-levelfactor. For top-down effects, this latter strategy is oftenmore practical because it involves positively replicatingthe relevant effect; in contrast, the former strategy mayrequire a null effect (which raises familiar concerns aboutstatistical power, etc.). In either case, however, such strat-egies show how this pitfall is eminently testable.

4.5. Pitfall 5: Peripheral attentional effects

We have been arguing that there are no top-down effects ofcognition on perception, in the strong and revolutionarysense wherein such effects violate informational encapsula-tion or cognitive impenetrability and so threaten the view ofthe visual system as a functionally independent (modular)part of the mind. However, we have also noted someother senses of top-down effects that carry no such implica-tions (see sect. 2). Chief among these is the notion ofchanging what we see by changing the input to perception,as when we close (or move) our eyes based on our desires(see sect. 2.1).Other ways of changing the input to perception,

however, are more subtle. Perhaps most prominently, shift-ing patterns of attention can change what we see. Selectiveattention is obviously closely linked to perception – oftenserving as a gateway to conscious awareness in the firstplace, such that we may completely fail to see what wedo not attend to (as in inattentional blindness; e.g., Mostet al. 2005b; Ward & Scholl 2015). Moreover, attention –which is often likened to a “spotlight” or “zoom lens” (seeCave & Bichot 1999; though cf. Scholl 2001) – can some-times literally highlight or enhance attended objects,making them appear (relative to unattended objects)clearer (Carrasco et al. 2004) and more finely detailed(Gobell & Carrasco 2005).Attentional phenomena relate to top-down effects

simply because attention is at least partly under inten-tional control – insofar as we can often choose to payattention to one object, event, feature, or region ratherthan another. When that happens – say, if we attend toa specific flower and it looks clearer or more detailed –should that not then count as our intentions changingwhat we see?In many such cases, changing what we see by selec-

tively attending to a different object or feature (e.g., topeople passing a basketball rather than to a dancinggorilla, or to black shapes rather than white shapes;Most et al. 2005b; Simons & Chabris 1999) seemsimportantly similar to changing what we see by movingour eyes (or turning the lights off). In both cases, weare changing the input to mechanisms of visual percep-tion, which may then still operate inflexibly given thatinput. A critical commonality, perhaps, is that the influ-ence of attention (or eye movements) in such cases iscompletely independent of your reason for attendingthat way. Having the lights turned off has the same




effect on visual perception regardless of why the lightsare off, including whether you turned them off inten-tionally or accidentally; in both cases it’s the change inthe light doing the work, not the antecedent intention.And in similar fashion, attention may enhance whatyou see regardless of the reasons that led you todeploy attention in that way, and even whether youattended voluntarily or through involuntary attentionalcapture; in both cases, it’s the change in attentiondoing the work, not the antecedent intention. Put differ-ently, such attentional (or light-turning-off) effects maybe occasioned by a relevant intention or belief, butthey are not sensitive to the content of that intentionor belief.Moreover, such attentional effects are already part of the

“orthodoxy” in vision science, which currently studies andmodels such attentional effects and readily accepts thatshifts in attention can affect what we see. By contrast, aprimary factor that makes other top-down effects (e.g.,effects of morality, hunger, language, etc. on perception)potentially revolutionary in the present context is preciselythat they are not part of this traditional understanding ofvisual perception.Of course, not all attentional effects must be so periph-

eral in nature. In other contexts, attention may interact inrich and nuanced ways with unconscious visual representa-tions to effectively mold and choose a “winning” percept –changing the content of perception rather than merelyinfluencing what we focus on. (For an elaboration of howsuch attentional dynamics may interact with issues of mod-ularity, see Clark 2013). However, our contention in thispitfall is that the merely peripheral sorts of attention –involving simple changes in which locations, features, orobjects we focus on – can account for a wide variety ofapparent top-down effects on perception. As a result, wefocus on such peripheral forms of attention in the rest ofthis section, while not denying that attention can also inter-act with perception in much richer ways as well.In light of such considerations, it seems especially impor-

tant to determine for any alleged top-down state (e.g., anintention, emotion, or desire) whether that state is influenc-ing what we see directly (in which case it may underminethe view that perception is a functionally independentmodule) or whether it is (merely) doing so indirectly bychanging how we attend to a stimulus in relatively periph-eral ways – in which case it may simply change the input tovisual processing but not how that processing operates.

4.5.1. Case studies. Attention has a curious status in thelong-running debate about top-down effects. On the onehand, perhaps based on its prominence in previous discus-sions (especially Pylyshyn 1999), the kinds of thoughtsnoted in the previous section are almost always recognizedand accepted in most modern discussions of top-downeffects – including recent discussions reaching very differ-ent conclusions than our own (e.g., two of the mostrecent literature reviews of top-down effects, which con-cluded that top-do

Documents

Cognition does not affect perception: Evaluating the evidence for … · 2017. 5. 15. · Cognition does not affect perception: Evaluating the evidence for “top-down” effects